Is it time to use real-world data from primary care in Alzheimer’s disease?

Alzheimer’s disease (AD) is the most common form of dementia, a major worldwide public health concern, and a huge burden to patients, families, and public health systems. The lack of any effective treatment to tackle it and the aging of the population—age is the main risk factor for AD—may partly explain the forecasted strong increase in the worldwide prevalence of AD expected for the next decades [1]. However, some recent studies have shown a decline in the incidence or prevalence of AD in high-income countries, and improved education levels and better management of cardiovascular risk factors have been suggested as driving factors for such declining pattern [2, 3]. Therefore, updated regional or national epidemiology of AD is urged to adapt public policies to the secular and geographical trends of the disease [4]. Nevertheless, updated estimates of AD prevalence and incidence in Europe are scarce [5]; for example, epidemiological studies in Spain are based on data collected in face-to-face surveys conducted in the 1990s or 2000s [6‐10]. Real-world data refers to observational data routinely collected in healthcare settings derived from multiple sources: electronic health records (EHR), disease registries, claims and billing data, and even data gathered through personal devices and health applications. Real-world data collected in primary care settings might provide a new opportunity to study the epidemiology of AD; general practitioners, as gatekeepers of the healthcare services, might provide a large number of cases compared to other clinical databases. However, epidemiological studies based on real-world data from primary care have hardly addressed specific dementia subtypes, such as AD [11‐13]. This lack of evidence may be related to the high level of under-diagnosis of dementia or AD cases in primary care [14]. However, recent reports stated good accuracy of diagnoses of dementia and AD recorded in primary care databases for research purposes [15‐17]. Therefore, we hypothesized that real-world data could provide regular updates on AD prevalence and incidence estimates, at reduced economical costs.

We sought to estimate the crude and standardized prevalence and incidence rates of AD in Catalonia using data from a large primary care database, including stratified results by age and sex. Additionally, we aimed to examine the suitability of real-world data to study the epidemiology of AD, comparing our estimates with those from previous face-to-face studies.

This is a population-based study conducted in Catalonia using a large EHR database from primary care.

This is the first comprehensive epidemiological study on AD to update prevalence and incidence estimates based on real-world data from primary care. In this respect, we contribute to fulfill a concerning gap in the literature and aim to delve into the epidemiology of AD using real-world data from primary care. Our findings are strengthened by the previous validation of the algorithm used to identify AD cases [17].

Real-word data are increasingly used in overall dementia research [16, 35, 36], but not in the field of AD, where it is surprisingly incipient. To our knowledge, the literature about AD epidemiology based on EHR is limited and restricted to the British population. We only found two studies that used primary care databases to estimate the incidence rate of AD [11, 37], and one study based on EHR from hospitals, which estimated AD prevalence in 5% and incidence rate in 3.7 per 1000 person-years in people aged at least 75 years [38]. Our estimates were higher than the ones reported by the abovementioned study based on EHR from hospitals, since we obtained an AD prevalence of 6.7% and an incidence of 8.1 per 1000 person-years among people aged at least 75 years old. Estimates from primary care databases are expected to be higher than those from secondary care, because there is a time frame in which the diagnosis might only be in the primary care records—if the patient has not yet been admitted to hospital. Surprisingly, AD incidence estimates from previous studies based on primary care databases were more similar to the study based on hospital data than to our estimates of 4.2 per 1000 person-years (95%CI 3.8–4.6). The AD incidence rate was estimated in 1.5 per 1000 person-years using data from The Health Improvement Network database (THIN) [37] and in 1.59 per 1000 person-years using the General Practice Research Database (GPRD) [11], two databases that contain EHR from primary care services in the UK. Our estimate of AD incidence rate (4.2 per 1000 person-years (95%CI 3.8–4.6)) was higher than those in previous studies based on EHR from primary care [11, 37, 38].

Unfortunately, we could not compare our prevalence of AD with similar literature because, to the best of our knowledge, no previous studies based on EHR from primary care have reported such estimates. Other studies were based on EHR, but did not assess AD, only all-cause dementia. For example, Jaakkimainen et al. [39] estimated the prevalence of overall dementia in 72.0 per 1000 persons using EHR recorded by family physicians from Canada. Perera et al. [35] reported prevalences of overall dementia in 2012 of 0.4, 1, 2, 5, 8, 10, and 11 for the following age groups: 65 to 70, 70 to 75, 75 to 80, 80 to 85, 85 to 90, 90 to 95, and ≥ 95 years, respectively. Since AD corresponds to about 50–60% of all dementia cases, we expected to find lower prevalences than the ones observed by Jaakkimainen et al. [39] or by Perera et al. [35], but we obtained estimates similar or slightly lower than the abovementioned prevalences of overall dementia. This could suggest that SIDIAP might capture a higher number of AD and dementia cases than databases from other healthcare systems [35].

The literature on AD epidemiology from face-to-face studies is much more comprehensive than that from EHR studies. Our incidence figures were concordant with the Delphi consensus study that reported a dementia incidence rate of 8.8 per 1000 person-years in Central Europe [1], about 60% of which corresponded to AD [9]. However, our incidence estimates were lower than the observed in two meta-analyses that provided estimates of 15.8 per 1000 person-years worldwide [40] and 11.1 per 1000 person-years in Europe [41]. Notably, both meta-analyses reported significantly high heterogeneity and thus should be interpreted cautiously. In Spain, previous face-to-face studies estimated AD incidence rates of 7.4 per 1000 person-years (95%CI 6.0–8.8) [10] and 10.8 per 1000 person-years (95%CI 7.8–13.7) in people aged at least 65 and 75 years, respectively [42], while our crude estimates were 4.7 (95%CI 4.6–4.8) and 8.09 (95%CI 7.84–8.34) per 1000 person-years in persons aged at least 65 and 75 years, respectively. Our prevalence estimates were slightly lower than the ones reported by face-to-face studies. Age-standardized AD prevalence was estimated in 4.2% worldwide [40], and in 4.4% or 5.1% in Europe [41, 43], while our estimate was 3.3% (95%CI 3.1–3.5). In Spain, age- and sex-standardized prevalence was estimated in 4.0% in people aged ≥ 65 years and in 5.6% in people aged ≥ 75 in face-to-face studies [6, 7]; these findings were slightly lower and similar to our estimates of 3.1% (95%CI 2.7–3.6) in the population aged ≥ 65 years and 5.8% (5.3–6.4) in the population aged ≥ 75 years. Our crude prevalence estimate (3.8% (95%CI 3.7–3.8)) fell within the low range of prevalence estimates from previous studies in the Spanish population, which was from 2.9 to 6.9% [44‐47].

Our estimates of prevalence and incidence of AD were similar or slightly lower than face-to-face studies, which could indicate a decreasing trend in AD incidence and prevalence. Previous face-to-face studies were conducted in Spain during the 1990s, and thenceforth, high-income countries have presented a decreasing trend in dementia [3] or AD [48] incidence. However, other countries only observed a decreasing trend in vascular dementia but not in AD [49]; the incidence of AD might actually be increasing, and the decline observed in the incidence of dementia in general could be related to a decline in vascular dementia. Further research to clarify the trend of AD and of vascular dementia would be of interest. Our lower estimates could also be explained by methodological differences, such as population characteristics, different case definitions, or under-registration of AD. In our study, under-registration of AD cases might have occurred for different reasons. First, under-registration is expected to be higher in studies based on EHR than in face-to-face studies, especially in the early stages of AD. In face-to-face studies, patients affected by subclinical symptoms or early stages of AD might likely be identified as AD cases because they are evaluated by specialists during the study. But in studies based on EHR, it is possible that many of these patients have not been cognitively assessed by the end of the study period, and therefore, they are not identified as AD cases. Second, the analysis of free text has been reported to enhance the identification of dementia cases [15]. But we used an algorithm that identified AD cases exclusively based on codes. Therefore, those AD diagnoses recorded in the clinical history using the free text instead of the ICD-10 codes enhanced under-registration in our study. Third, patients diagnosed by a specialist from private healthcare might not have the AD diagnosis recorded in their clinical history, unless they (or their relatives) inform the general practitioner about their condition. Despite the universal coverage of the Catalan public health system, about 25% of persons aged at least 60 years have contracted a private health insurance. However, not all of them might have contributed to under-registration in our study. Most of the private healthcare users also consult public healthcare professionals, especially among the elderly. The public health system provides free access or low copayment for pharmacological treatment prescribed by general practitioners to all Catalan pensioners. In our study, we identified AD cases combining data recorded in the primary care settings with data from the pharmacy-invoicing database. Therefore, we could identify as AD cases those patients attended by a private specialist who asked their general practitioner for prescription of anti-dementia drugs to benefit from economic discounts or for a prescription renewal.

In SIDIAP, under-registration of AD was concentrated in the oldest group: we observed increasing trends of AD prevalence and incidence with age except for this group (≥ 90 years), in line with previous studies [50]. Some general practitioners might be reluctant to use dementia codes among the oldest persons because they may consider the diagnosis and treatment as ineffectual in this population or may interpret their memory difficulties as part of the normal aging process rather than as a disability [16]. Besides, people aged at least 90 years might have higher levels of comorbidity that could affect the quality of AD diagnosis and register. For example, those persons who have impaired mobility due to chronic complex diseases might not be able to visit the specialist, resulting in a diagnosis of unspecified dementia instead of AD; this is in line with previous findings that reported the highest prevalence and incidence rate estimates of overall dementia among the oldest elderly in Catalonia [16]. Patients institutionalized in private geriatrics or residences might lose contact with their general practitioner, increasing under-registration of EHR. Moreover, patients with vascular comorbidities who had a diagnosis of AD and were diagnosed with another dementia subtype (such as mixed dementia or vascular dementia) were excluded from our study, because the algorithm we used only defined dementia cases as AD if this was the last diagnosis. Finally, certain aspects other than under-registration could partially explain our lower estimates in the oldest elderly: some level of survival bias that could not be rejected and an attenuation of the relationship between AD diagnosis and clinical expression of dementia in the oldest old, which has been pointed out in recent studies [49].

Our estimates were clearly higher than those in previous studies based on EHR, suggesting that under-diagnosis in primary care settings probably varies strongly depending on the clinical practice. In the Catalan primary care, underestimation appeared to be less frequent than in other primary care settings, which reported that only half of the expected number of patients with dementia were recognized by the general practitioner [14]. This could be partly explained because the accuracy of AD diagnosis routinely collected in primary care might improve in active dementia diagnosis centers; in our study settings, about 25% of patients without dementia and 75% of patients with dementia aged 65 years or over have taken a cognitive test [16]. Additionally, about 90% of the AD prevalent cases recorded in primary care had been evaluated by a specialist, which suggests that general practitioners are coordinated with the secondary care settings [17]. This coordination is promoted by initiatives like an asynchronous telemedicine program that establishes protocols for the screening and diagnosis of dementia, thereby enabling primary care and specialized care professionals to reach shared diagnoses of dementia [51]. Finally, under-registration in our study was probably moderate because we might have captured a high number of AD cases among treated patients. In Catalonia, about 68% of dementia patients receive anti-dementia drugs [52], and this percentage may be higher than in other countries: in France, it is about 38% [53], and in Germany, it is about 52% [54]. Furthermore, our definition of AD case might have contributed to increase the number of AD cases identified by anti-dementia treatment. We included patients treated with anti-dementia drugs at any time—that is, patients who were treated before the study period were also included—whereas other studies accounted for currently treated patients—patients who were medicated in the past were excluded.

We acknowledge some limitations. First, our results could be underestimated because we did not use free text recorded in the clinical history, an accredited method to identify persons with AD [15]. However, we used a validated algorithm to identify AD cases [17] that combined diagnosis with prescription or dispensing codes, involving not only data from general practitioners but also from a pharmacy-invoicing database. In Catalonia, the prescription of anti-dementia drugs is requested by the general practitioner but requires approval by certain other specialists (geriatricians, psychiatrists, or neurologists). The use of two different data sources minimized the possible under-recording of dementia diagnoses in primary care. Second, the use of certain additional clinical parameters to diagnose AD could have increased the number of identified AD cases. However, we prioritized the use of a simple algorithm that was validated and had a good positive predictive value (72.3%, 95%CI 70.7–73.9), and sensitivity (83.3, 95%CI 81.8–84.6) [17]. Third, we did not account for educational level, a risk factor for AD, because such information was not available in SIDIAP. Fourth, AD is one of the many dementia disorders, and epidemiological trends might vary between dementia subtypes, for example, decreasing trends of overall dementia have been reported due to the reduction of vascular dementia but not of AD [49]. Thus, further details on the epidemiology of other dementia subtypes are crucial to plan effective preventive measures against dementia.

The main strength of our study is our large sample size (including more than 1 million persons aged at least 65 years old, and 40,000 dementia cases), which enhanced external validity; SIDIAP contains high-quality data from about 80% of the Catalan population and good representativeness in terms of geographical, age, and sex distributions [25].

The prevalence and incidence rates of AD based on EHR from primary care were higher than other estimates based on primary care databases from other European countries and slightly lower than the most recent face-to-face studies. Despite some limitations, real-world data routinely collected in primary care settings could be applied to epidemiological studies to complement face-to-face studies and could provide regular updates of prevalence and incidence estimates of AD at a reasonable cost. Better monitoring of the AD trends might help adjust the national dementia plans and the allocation of resources for families and caregivers to the current number of cases occurring in the community. Additionally, our results contribute to raise awareness about the importance of the accuracy of EHR and might encourage general practitioners, nurses, and specialists to keep improving the quality of the data they routinely collect in the electronic health histories. It is time to use real-world data from the primary care settings to update AD epidemiology complementing face-to-face studies, which might contribute to improve dementia policies and resource management for patients and their families.